fix(insights): add logging to diagnose why AI network observations never persist

The parse function silently returned {:ok, []} on any decode failure or
observation rejection. Log warnings for both cases so we can see what the
LLM is actually returning.

lib/towerops/llm/network_insight_prompt.ex

@@ -10,6 +10,8 @@ defmodule Towerops.LLM.NetworkInsightPrompt do
   comments on findings the rule-based workers already produced.
   """
 
+  require Logger
+
   @valid_urgencies ~w(critical warning info)
 
   @system """
@@ -71,12 +73,21 @@ defmodule Towerops.LLM.NetworkInsightPrompt do
     observations =
       case decode(cleaned) do
         {:ok, %{"observations" => list}} when is_list(list) ->
-          list
-          |> Enum.map(&normalize_observation/1)
-          |> Enum.reject(&is_nil/1)
-          |> Enum.take(@max_observations)
+          raw_count = length(list)
+          normalized = list |> Enum.map(&normalize_observation/1) |> Enum.reject(&is_nil/1)
+          rejected = raw_count - length(normalized)
+
+          if rejected > 0 do
+            Logger.warning("AI network insight: #{rejected}/#{raw_count} observations rejected by normalize_observation")
+          end
+
+          Enum.take(normalized, @max_observations)
 
         _ ->
+          Logger.warning(
+            "AI network insight: could not decode LLM response, content preview: #{String.slice(cleaned, 0, 200)}"
+          )
+
           []
       end